High-quality labeled data is essential for machine learning, particularly with auditory information, where events must be precisely annotated by start and end times. To ensure reliability, multiple annotators are often involved, but discrepancies are common. We present the Software Visual Tool for Audio Annotations, designed to support annotators by visually highlighting mismatches in event boundaries and facilitating consensus. Additionally, the tool integrates Cohen’s Kappa to quantify inter-annotator agreement, offering valuable feedback to refine annotation processes. This approach contributes to the creation of more reliable, high-quality auditory datasets for machine learning applications. • Visual tool detects and highlights mismatched audio event boundaries. • Integrates Cohen’s Kappa to measure inter-annotator agreement in real time. • Enhances reliability of labeled auditory datasets for machine learning.
Carballo et al. (Tue,) studied this question.